multivariate-probit 0.3.0__tar.gz → 0.3.2__tar.gz

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/CITATION.cff

@@ -7,7 +7,7 @@ authors:
     given-names: Mark
     orcid: "https://orcid.org/0009-0008-7193-1263"
 version: 0.3.0
-date-released: 2026-10-04
+date-released: 2026-10-06
 license: MIT
 repository-code: "https://github.com/sign-of-fourier/multivariate-probit"
 url: "https://quantecarlo.com/multivariate-probit"

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: multivariate-probit
-Version: 0.3.0
+Version: 0.3.2
 Summary: Joint probabilities for correlated binary outcomes: P(any), P(all), patterns and conditionals that agree. Any classifier per outcome; scikit-learn-style API.
 Project-URL: Homepage, https://quantecarlo.com/multivariate-probit
 Project-URL: Source, https://github.com/sign-of-fourier/multivariate-probit
@@ -213,8 +213,10 @@ The backend is chosen with `evaluator=`:
 of magnitude faster than SciPy, at roughly the accuracy of 100-draw GHK
 simulation ([studies/ghk.md](https://github.com/sign-of-fourier/multivariate-probit/blob/main/docs/studies/ghk.md)). It ships for CPython 3.11
 and 3.12 on x86-64 Linux (`pip install multivariate-probit[orthant]`); without
-a key it runs d ≤ 3 at `resolution="low"`. Key:
-[quantecarlo.com/orthant_key](https://quantecarlo.com/orthant_key).
+a key it runs d ≤ 3 at `resolution="low"`. The key is read from `$ORTHANT_KEY` or
+`~/.orthant/key` at import, or passed in code with
+`from multivariate_probit import orthant; orthant.activate(key)` (on Spark, call it inside
+the function the workers run). Key: [quantecarlo.com/orthant_key](https://quantecarlo.com/orthant_key).
 
 Benchmarks against other tools: [docs/benchmarks.md](https://github.com/sign-of-fourier/multivariate-probit/blob/main/docs/benchmarks.md).
 

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/README.md

@@ -149,8 +149,10 @@ The backend is chosen with `evaluator=`:
 of magnitude faster than SciPy, at roughly the accuracy of 100-draw GHK
 simulation ([studies/ghk.md](https://github.com/sign-of-fourier/multivariate-probit/blob/main/docs/studies/ghk.md)). It ships for CPython 3.11
 and 3.12 on x86-64 Linux (`pip install multivariate-probit[orthant]`); without
-a key it runs d ≤ 3 at `resolution="low"`. Key:
-[quantecarlo.com/orthant_key](https://quantecarlo.com/orthant_key).
+a key it runs d ≤ 3 at `resolution="low"`. The key is read from `$ORTHANT_KEY` or
+`~/.orthant/key` at import, or passed in code with
+`from multivariate_probit import orthant; orthant.activate(key)` (on Spark, call it inside
+the function the workers run). Key: [quantecarlo.com/orthant_key](https://quantecarlo.com/orthant_key).
 
 Benchmarks against other tools: [docs/benchmarks.md](https://github.com/sign-of-fourier/multivariate-probit/blob/main/docs/benchmarks.md).
 

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/benchmarks/bench.py

@@ -104,8 +104,8 @@ def run_cell(d, n_train=3000, n_test=100, rho=0.4, seed=0, sigma="equi"):
     if d <= 6:
         res["mvp quadrature"] = timed(lambda: model.set_params(evaluator="quadrature").joint_proba(Xte, Yte))
     if orthant is not None and (ORTHANT_KEYED or d <= 3):
-        r = "high" if ORTHANT_KEYED else "low"
-        res[f"mvp orthant ({r})"] = timed(lambda: model.set_params(evaluator="orthant", resolution=r).joint_proba(Xte, Yte))
+        for r in (("high", "highest") if ORTHANT_KEYED else ("low",)):
+            res[f"mvp orthant ({r})"] = timed(lambda: model.set_params(evaluator="orthant", resolution=r).joint_proba(Xte, Yte))
     res["SciPy default"] = timed(lambda: model.set_params(evaluator="scipy").joint_proba(Xte, Yte))
     if mvncd_batch is not None:
         st = np.moveaxis(stack, 2, 0)

multivariate_probit-0.3.2/benchmarks/results.csv

@@ -0,0 +1,65 @@
+sigma,d,N,method,median_err,max_err,s_per_row
+equi,3,100,mvp quadrature,1.3373382605444406e-05,5.564767189014663e-05,7.751391967758536e-05
+equi,3,100,mvp orthant (high),5.190348623980261e-05,0.00019774969627306405,1.484813867136836e-05
+equi,3,100,mvp orthant (highest),1.999845194528005e-06,7.989768001565034e-06,4.311794182285666e-05
+equi,3,100,SciPy default,2.039637077871914e-06,1.1415786741186817e-05,0.005793280370999128
+equi,3,100,pybhatlib ovus,0.00011229346488104974,0.0008524556947959311,8.469653083011508e-05
+equi,3,100,statsmodels,0.028443441360488153,0.09919722660040897,6.613903678953648e-07
+equi,3,100,reference,0.0,0.0,0.2536748156798421
+equi,5,100,mvp quadrature,2.0314513939401896e-05,0.00011324180994054522,0.04333823672030121
+equi,5,100,mvp orthant (high),8.080162918654848e-05,0.00030384074452138865,2.3601381108164788e-05
+equi,5,100,mvp orthant (highest),3.4247200694226265e-06,2.3859122737224236e-05,7.270404836162925e-05
+equi,5,100,SciPy default,1.3427915335623153e-06,7.990427117937626e-06,0.010302326420787721
+equi,5,100,pybhatlib ovus,0.00039917378471633624,0.0019016346577496457,0.00011997280875220894
+equi,5,100,statsmodels,0.024622620503874038,0.12870241315304662,5.677388980984688e-07
+equi,5,100,reference,0.0,0.0,0.5339089114402304
+equi,8,100,mvp orthant (high),7.013787736394328e-05,0.00045342350585386715,4.317081067711115e-05
+equi,8,100,mvp orthant (highest),4.605751738051556e-06,2.9230892634346528e-05,9.955347049981356e-05
+equi,8,100,SciPy default,1.0030036643121978e-06,8.134689687355756e-06,0.01642337770899758
+equi,8,100,pybhatlib ovus,0.0003370945286850363,0.0023083975100191556,0.0001771104708313942
+equi,8,100,statsmodels,0.014453125416131472,0.11538392212916126,9.872694499790668e-07
+equi,8,100,reference,0.0,0.0,1.4400106923800196
+equi,12,100,mvp orthant (high),6.245144142705012e-05,0.0007179989696481925,6.305043119937181e-05
+equi,12,100,mvp orthant (highest),5.63448417025339e-06,4.668168500868153e-05,0.00014906150056049228
+equi,12,100,SciPy default,3.440844550360589e-07,7.96331593577726e-06,0.02377823827089742
+equi,12,100,pybhatlib ovus,0.0002047092197251764,0.0035788660943063266,0.0002432926092296839
+equi,12,100,statsmodels,0.006303004648829415,0.07762653255283464,1.5006307512521743e-06
+equi,12,100,reference,0.0,0.0,1.6423227695701643
+equi,16,100,mvp orthant (high),1.630594965073808e-05,0.0006243813016577615,9.883297141641379e-05
+equi,16,100,mvp orthant (highest),1.1074220600440065e-06,2.9310608489277468e-05,0.00022130653960630297
+equi,16,100,SciPy default,6.853848620851782e-08,4.3920697129634645e-06,0.03263484545052051
+equi,16,100,pybhatlib ovus,1.6800734753088424e-05,0.0011253768512156778,0.00031500173965469
+equi,16,100,statsmodels,0.0014514737962008938,0.05355898580397629,1.4994502998888493e-06
+equi,16,100,reference,0.0,0.0,1.1454083026899025
+pairwise,3,100,mvp quadrature,4.975128970532983e-06,8.276622284841828e-05,7.341289892792702e-05
+pairwise,3,100,mvp orthant (high),0.0006363036604397354,0.002893273763736187,2.0136469975113868e-05
+pairwise,3,100,mvp orthant (highest),0.00011323696696274865,0.0007099285566469216,4.4253640808165074e-05
+pairwise,3,100,SciPy default,1.3494787358958593e-06,2.586364471635072e-05,0.0073384556197561325
+pairwise,3,100,pybhatlib ovus,0.00037654161082817383,0.004713394298326956,8.070134790614248e-05
+pairwise,3,100,statsmodels,0.037060543222106224,0.15997644750429899,4.6383123844861987e-07
+pairwise,3,100,reference,0.0,0.0,0.24052267621998907
+pairwise,5,100,mvp quadrature,5.607947987679163e-06,6.266772635254991e-05,0.04593952419003472
+pairwise,5,100,mvp orthant (high),0.0060661393002539865,0.03370856471106998,2.4112039245665073e-05
+pairwise,5,100,mvp orthant (highest),0.003141591472459862,0.0325453719435822,6.538510089740157e-05
+pairwise,5,100,SciPy default,1.2512113251161167e-06,2.1201235567813725e-05,0.14286554857157172
+pairwise,5,100,pybhatlib ovus,0.0014663128072703105,0.02764332982515838,0.00012095558922737837
+pairwise,5,100,statsmodels,0.052292238612919734,0.2549403651447862,5.610100924968719e-07
+pairwise,5,100,reference,0.0,0.0,1.2560263323498657
+pairwise,8,100,mvp orthant (high),0.0047540425543466,0.03786870554320598,3.998839994892478e-05
+pairwise,8,100,mvp orthant (highest),0.0021531855423506108,0.013414597153998009,9.109209990128875e-05
+pairwise,8,100,SciPy default,2.2860621589921443e-06,3.321167501391109e-05,1.132777205838356
+pairwise,8,100,pybhatlib ovus,0.0015355099588129813,0.010313908094296734,0.00017519862158223986
+pairwise,8,100,statsmodels,0.024908203618945243,0.1426044668035415,8.142902515828609e-07
+pairwise,8,100,reference,0.0,0.0,4.434727118879964
+pairwise,12,100,mvp orthant (high),0.003428055959417961,0.04309079684926198,6.395480129867793e-05
+pairwise,12,100,mvp orthant (highest),0.002445889327158528,0.02153141541464762,0.00015708803897723555
+pairwise,12,100,SciPy default,1.9157011284364617e-06,5.195613406111521e-05,1.5738827282516286
+pairwise,12,100,pybhatlib ovus,0.0007431598443556342,0.01388904489935694,0.0002463792893104255
+pairwise,12,100,statsmodels,0.014263734107914516,0.1854686635860365,1.2842798605561255e-06
+pairwise,12,100,reference,0.0,0.0,9.518902401560045
+pairwise,16,100,mvp orthant (high),0.0037733872849075424,0.04512852480876324,8.360587060451507e-05
+pairwise,16,100,mvp orthant (highest),0.00293222915103126,0.045234964595115376,0.00021089559188112618
+pairwise,16,100,SciPy default,1.7024296601283496e-06,1.0376357636343592e-05,1.7738397101382726
+pairwise,16,100,pybhatlib ovus,0.0009786447220386783,0.015062755675490355,0.0003229634207673371
+pairwise,16,100,statsmodels,0.010651175036494289,0.18479811475521019,1.8178089521825314e-06
+pairwise,16,100,reference,0.0,0.0,19.51646036452992

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/docs/api.md

@@ -36,7 +36,7 @@ MultivariateProbit(
 | `project_correlation` | `True` | Project a pairwise estimate onto the nearest positive-definite correlation matrix. No effect when `dependence="joint"`. |
 | `random_state` | `None` | Controls the cross-fitting split and `sample`. |
 | `evaluator` | `"quadrature"` | Backend for the orthant probabilities behind `dependence="joint"` and every joint query: `"quadrature"`, `"scipy"` or `"orthant"`. See [Evaluators](#evaluators). |
-| `resolution` | `"high"` | Passed to `evaluator="orthant"`; ignored otherwise. |
+| `resolution` | `"high"` | Passed to `evaluator="orthant"`; ignored otherwise. `"high"`, `"highest"` (most accurate, about 2–3× the time of `"high"`: median error under 0.00005 on an equicorrelated Σ against up to 0.0001 at `"high"`, and 0.0001–0.0031 on a near-singular Σ against 0.0006–0.0061; see [benchmarks.md](benchmarks.md)), `"smooth"` (the cost of `"highest"`, continuous in the parameters with no threshold where the method changes, for gradient-based fitting; not uniformly more accurate than `"high"`) or `"low"` (cheapest). |
 | `fitter` | `"ifm"` | `"ifm"`: the local two-stage fit. `"modal"`: full maximum likelihood on a hosted GPU service ([fiml.md](fiml.md)) — `inner="linear"` only, `inner_params` may set only `alpha`; `dependence`, `cv`, `optimizer` and `project_correlation` do not apply. Uploads X, Y and `sample_weight`. An int `random_state` seeds its Sobol draws (default seed 0). |
 | `n_draws` | `512` | Sobol draws per row for `fitter="modal"`; ignored otherwise. |
 
@@ -77,7 +77,7 @@ queries on a pairwise fit still use the chosen backend.
 | --- | --- | --- |
 | `"quadrature"` | Genz's recursive conditioning with a fixed Gauss-Legendre rule of order `n_quad` (see [implementation.md](implementation.md)). | Deterministic and accurate. Cost grows as `n_quad ** (d - 2)`, so it becomes impractical well before d = 10. |
 | `"scipy"` | `scipy.stats.multivariate_normal.cdf`, one row at a time. | Reaches any d, but is slow per row and randomised: a `dependence="joint"` fit is not exactly reproducible, and its noisy objective can stall the simplex. |
-| `"orthant"` | An optional compiled package bundled as `multivariate_probit.orthant`. Install with `pip install multivariate-probit[orthant]`. | Built for CPython 3.11 and 3.12 on x86-64 Linux only; elsewhere it raises `ImportError`. Without a key it accepts d ≤ 3 at `resolution="low"` only, and raises `ValueError` outside that. A key is read from `$ORTHANT_KEY` or `~/.orthant/key`; see https://quantecarlo.com/orthant_key. Accuracy and speed against GHK and SciPy: [studies/ghk.md](studies/ghk.md). |
+| `"orthant"` | An optional compiled package bundled as `multivariate_probit.orthant`. Install with `pip install multivariate-probit[orthant]`. | Built for CPython 3.11 and 3.12 on x86-64 Linux only; elsewhere it raises `ImportError`. Without a key it accepts d ≤ 3 at `resolution="low"` only, and raises `ValueError` outside that. A key is read from `$ORTHANT_KEY` or `~/.orthant/key` at import, or passed at any time with `orthant.activate(key)` (returns `"paid"`; a wrong key raises `ValueError`). Each process needs its own key: on Spark, call `activate` inside the function the workers run, or set the variable in the cluster environment; see https://quantecarlo.com/orthant_key. Accuracy and speed against GHK and SciPy: [studies/ghk.md](studies/ghk.md). |
 
 For test sets too large for one machine, the same evaluator runs as a hosted
 GPU service, `quantecarlo.orthant_cdf`: about a second per million rows at
@@ -94,7 +94,7 @@ the
 | `correlation_` | (d, d) | The fitted Σ. |
 | `eta_` | (n, d) | The (cross-fitted) latent indices stage two was fitted on; for `fitter="modal"`, the in-sample indices of the fitted model. |
 | `calibration_` | (d, 2) | Intercept and slope of a probit of each outcome on its own fitted index — the calibration slope. See the note below. |
-| `nll_` | float | Negative log-likelihood at the end of the dependence fit. Joint and pairwise fits optimise different objectives, so the values are not comparable across settings. |
+| `nll_` | float | Negative log-likelihood at the end of the dependence fit. Joint and pairwise fits optimise different objectives, so the values are not comparable across settings. For `fitter="modal"`, the simulated joint negative log-likelihood of all margins and Σ. |
 | `optimize_result_` | OptimizeResult or None | The SciPy result for `dependence="joint"`. `None` for `fitter="modal"`. |
 | `stderr_`, `stderr_robust_` | dict | `fitter="modal"` only. Standard errors keyed `"coef"` (d, n_features), `"intercept"` (d,), `"correlation"` (d, d); inverse-Hessian and sandwich. |
 | `cov_params_`, `cov_params_robust_` | (k, k) | `fitter="modal"` only. The matching covariance matrices, ordered as `param_names_`. |

multivariate_probit-0.3.2/docs/benchmarks.md

@@ -0,0 +1,95 @@
+# Benchmarks against other tools
+
+**Task.** Score `P(Y = y | x)`, the probability of the pattern each held-out
+row actually has, from one fitted model. Synthetic data: 6 features, linear
+margins, 3,000 training rows, 100 scored rows per d. The model is
+`MultivariateProbit(inner="linear", dependence="pairwise", random_state=0)`.
+Every column scores that same fitted model (η̂, Σ̂), except statsmodels.
+
+There are two shapes of true Σ:
+
+- **Equicorrelated**, ρ = 0.4: the easy case.
+- **Pairwise-shaped**: a rank-2 factor plus noise, projected to the nearest
+  correlation matrix (smallest eigenvalue about 1e-8), the same construction
+  as [studies/ghk.md](studies/ghk.md). A fitted pairwise Σ̂ lands on that PSD
+  boundary, which is hard for every approximation.
+
+**Reference.** SciPy `multivariate_normal.cdf` with `abseps=1e-7`,
+`releps=1e-6`, one row at a time. Error is the absolute difference from it in
+probability: median over rows, maximum in parentheses. Time is wall-clock per
+row after one warm-up call, on a two-core x86-64 Linux machine.
+
+Reproduce with [benchmarks/benchmarks.ipynb](../benchmarks/benchmarks.ipynb)
+(logic in `benchmarks/bench.py`; raw numbers in `benchmarks/results.csv`).
+The reference values are cached in `benchmarks/ref_*.npz` together with the
+problem they were computed for, so a rerun skips the slow part and a changed
+problem recomputes. The `orthant` column needs a key in `$ORTHANT_KEY` or
+`~/.orthant/key`.
+
+### Equicorrelated Σ (ρ = 0.4)
+
+| d | median p | multivariate-probit `orthant` (`resolution="high"`) | multivariate-probit `orthant` (`resolution="highest"`) | multivariate-probit `quadrature` | SciPy, default tolerances | pybhatlib (Bhat OVUS) | statsmodels |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| 3 | 0.348 | 1.5e-05 s / 0.0001 (0.0002) | 4.3e-05 s / 0.0000 (0.0000) | 7.8e-05 s / 0.0000 (0.0001) | 5.8e-03 s / 0.0000 (0.0000) | 8.5e-05 s / 0.0001 (0.0009) | 6.6e-07 s / 0.0284 (0.0992) |
+| 5 | 0.154 | 2.4e-05 s / 0.0001 (0.0003) | 7.3e-05 s / 0.0000 (0.0000) | 4.3e-02 s / 0.0000 (0.0001) | 1.0e-02 s / 0.0000 (0.0000) | 1.2e-04 s / 0.0004 (0.0019) | 5.7e-07 s / 0.0246 (0.1287) |
+| 8 | 0.101 | 4.3e-05 s / 0.0001 (0.0005) | 1.0e-04 s / 0.0000 (0.0000) | impractical past d = 7 | 1.6e-02 s / 0.0000 (0.0000) | 1.8e-04 s / 0.0003 (0.0023) | 9.9e-07 s / 0.0145 (0.1154) |
+| 12 | 0.022 | 6.3e-05 s / 0.0001 (0.0007) | 1.5e-04 s / 0.0000 (0.0000) | impractical past d = 7 | 2.4e-02 s / 0.0000 (0.0000) | 2.4e-04 s / 0.0002 (0.0036) | 1.5e-06 s / 0.0063 (0.0776) |
+| 16 | 0.0024 | 9.9e-05 s / 0.0000 (0.0006) | 2.2e-04 s / 0.0000 (0.0000) | impractical past d = 7 | 3.3e-02 s / 0.0000 (0.0000) | 3.2e-04 s / 0.0000 (0.0011) | 1.5e-06 s / 0.0015 (0.0536) |
+
+### Pairwise-shaped Σ (near-singular)
+
+| d | median p | multivariate-probit `orthant` (`resolution="high"`) | multivariate-probit `orthant` (`resolution="highest"`) | multivariate-probit `quadrature` | SciPy, default tolerances | pybhatlib (Bhat OVUS) | statsmodels |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| 3 | 0.431 | 2.0e-05 s / 0.0006 (0.0029) | 4.4e-05 s / 0.0001 (0.0007) | 7.3e-05 s / 0.0000 (0.0001) | 7.3e-03 s / 0.0000 (0.0000) | 8.1e-05 s / 0.0004 (0.0047) | 4.6e-07 s / 0.0371 (0.1600) |
+| 5 | 0.171 | 2.4e-05 s / 0.0061 (0.0337) | 6.5e-05 s / 0.0031 (0.0325) | 4.6e-02 s / 0.0000 (0.0001) | 1.4e-01 s / 0.0000 (0.0000) | 1.2e-04 s / 0.0015 (0.0276) | 5.6e-07 s / 0.0523 (0.2549) |
+| 8 | 0.097 | 4.0e-05 s / 0.0048 (0.0379) | 9.1e-05 s / 0.0022 (0.0134) | impractical past d = 7 | 1.1e+00 s / 0.0000 (0.0000) | 1.8e-04 s / 0.0015 (0.0103) | 8.1e-07 s / 0.0249 (0.1426) |
+| 12 | 0.025 | 6.4e-05 s / 0.0034 (0.0431) | 1.6e-04 s / 0.0024 (0.0215) | impractical past d = 7 | 1.6e+00 s / 0.0000 (0.0001) | 2.5e-04 s / 0.0007 (0.0139) | 1.3e-06 s / 0.0143 (0.1855) |
+| 16 | 0.013 | 8.4e-05 s / 0.0038 (0.0451) | 2.1e-04 s / 0.0029 (0.0452) | impractical past d = 7 | 1.8e+00 s / 0.0000 (0.0000) | 3.2e-04 s / 0.0010 (0.0151) | 1.8e-06 s / 0.0107 (0.1848) |
+
+R `mvProbit` was not run (see below).
+
+The reference itself took 0.24 to 1.6 s per row on the equicorrelated Σ and
+0.24 to 20 s per row on the pairwise-shaped one. At d = 16 on the
+equicorrelated Σ most pattern probabilities are small (median 0.0024), so the
+maximum error says more than the median there.
+
+## Reading the tables
+
+- **statsmodels** fits one `Probit` per outcome and has no correlation matrix,
+  so its pattern probability is a product of marginals. Its error is the cost
+  of ignoring Σ, not of numerical integration, and it does not shrink with
+  more data. Its time covers prediction only.
+- **`orthant`** scores all rows in one compiled call.
+  - **`resolution="high"`** was the fastest of the integrating methods at every
+    d on both Σ shapes: 3 to 6 times faster than pybhatlib, and 330 to 440 times
+    faster than SciPy on the equicorrelated Σ. Median error 0.0001 or less on the
+    equicorrelated Σ (maximum up to 0.0007, against pybhatlib's 0.0009 to
+    0.0036). On the pairwise-shaped Σ, median error 0.0006 to 0.0061, maximum up
+    to 0.045, the largest of the integrating methods.
+  - **`resolution="highest"`** costs 2.2 to 3.1 times `"high"`'s time and is
+    still 1.4 to 2 times faster than pybhatlib. On the equicorrelated Σ its
+    median and maximum error are below 0.00005 at every d. On the
+    pairwise-shaped Σ its median error is 0.0001 to 0.0031 (`"high"`: 0.0006 to
+    0.0061) and its maximum up to 0.045; pybhatlib's median there is 0.0004 to
+    0.0015.
+- **pybhatlib** scores with Bhat's analytic approximation (`mvncd_batch`,
+  `method="ovus"`), one row at a time in Python. It is 70 to 105 times faster
+  than SciPy on the equicorrelated Σ. On the pairwise-shaped Σ its median error
+  is 0.0004 to 0.0015 and its maximum 0.0047 to 0.028. pybhatlib also has a PyTorch backend that runs on
+  your own GPU, but in pybhatlib 0.4.0 `mvncd` converts its inputs to NumPy
+  and `mvncd_batch` loops over rows in Python, so the GPU does not speed up
+  this scoring.
+- **SciPy** stays accurate on the pairwise-shaped Σ but slows sharply, to
+  1.1 to 1.8 s per row from d = 8. That is 54 to 69 times its time on the
+  equicorrelated Σ at the same d.
+- **`quadrature`** matched the reference on both shapes up to d = 5.
+- **R `mvProbit`** was not run: R is not installed on the benchmark machine,
+  and the notebook has no R column yet.
+
+## What this does not establish
+
+- Two Σ shapes, one seed, and 100 rows per d. The pairwise-shaped Σ is one
+  construction of a near-singular matrix; see [studies/ghk.md](studies/ghk.md)
+  and [studies/comparators.md](studies/comparators.md) for more.
+- Scoring only. Fitting time against full-likelihood estimation is in
+  [studies/comparators.md](studies/comparators.md).

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/docs/fiml.md

@@ -19,6 +19,27 @@ model.stderr_robust_          # sandwich standard errors
 model.predict_proba(X_new)    # runs locally, on any evaluator
 ```
 
+## What changes, and what doesn't
+
+Only `fitter="modal"` changes. After `fit`, the model is used exactly as an
+IFM fit is: `transform`, `predict_proba` and its joint queries, `joint_proba`,
+`score`, `sample`, `correlation_`, `inner_models_[j].coef_`, `get_params` /
+`set_params`. Prediction runs locally on the chosen `evaluator`.
+
+The differences:
+
+- `inner` must be `"linear"`; `inner_params` may set only `alpha`. Anything
+  else raises.
+- `dependence`, `cv`, `optimizer` and `project_correlation` do not apply.
+- `fit` uploads X, Y and `sample_weight` and waits for the service (see
+  [Cost](#cost)).
+- Extra attributes: `stderr_`, `stderr_robust_`, `cov_params_`,
+  `cov_params_robust_`, `param_names_`, `fit_result_`.
+- `eta_` holds in-sample indices (there is no cross-fitting); `nll_` is the
+  simulated joint negative log-likelihood; `optimize_result_` is `None`.
+- An int `random_state` seeds the Sobol draws; otherwise the seed is 0, so
+  repeated fits agree.
+
 ## When to use it instead of IFM
 
 - You need **standard errors** for the coefficients and for Σ. IFM does not

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/docs/studies/ghk.md

@@ -6,6 +6,11 @@ without measuring it. For *scoring* a fitted model, where that objection does
 not apply, how does GHK compare with the compiled `orthant` backend in
 accuracy and speed?
 
+> **Version note (2026-10).** This study measured the backend shipped in 0.2.1. Later builds integrate
+> `resolution="high"` on a finer rule and add `resolution="highest"`, which is more accurate at a few times the
+> cost. The tables below were not rerun; current numbers for both settings are in
+> [benchmarks.md](../benchmarks.md).
+
 ## Summary
 
 | | accuracy | speed (s per row) |

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "multivariate-probit"
-version = "0.3.0"
+version = "0.3.2"
 description = "Joint probabilities for correlated binary outcomes: P(any), P(all), patterns and conditionals that agree. Any classifier per outcome; scikit-learn-style API."
 readme = "README.md"
 license = { file = "LICENSE" }

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/src/multivariate_probit/__init__.py

@@ -34,7 +34,7 @@ from .linear import ProbitRegressor
 from .model import MultivariateProbit
 from .results import MultivariateProbitProba
 
-__version__ = "0.3.0"
+__version__ = "0.3.2"
 
 __all__ = [
     "MultivariateProbit",

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/src/multivariate_probit/model.py

@@ -165,8 +165,17 @@ class MultivariateProbit:
     n_quad : int, default=24
         Gauss-Legendre order for ``evaluator="quadrature"``. Lower it if
         fitting with many outcomes gets slow.
-    resolution : {"high", "low"}, default="high"
-        Passed to ``evaluator="orthant"``; ignored otherwise.
+    resolution : {"high", "highest", "smooth", "low"}, default="high"
+        Passed to ``evaluator="orthant"``; ignored otherwise. ``"highest"``
+        is the most accurate setting, at about 2-3 times the time of
+        ``"high"``: in docs/benchmarks.md its median error is under 0.00005 on
+        an equicorrelated covariance (``"high"``: up to 0.0001) and
+        0.0001-0.0031 on a near-singular one (``"high"``: 0.0006-0.0061).
+        ``"smooth"`` costs the same as ``"highest"`` and is continuous in the
+        parameters (no threshold where the method changes, near-duplicates
+        included), for gradient-based fitting; it is not uniformly more
+        accurate than ``"high"``. ``"low"`` is the cheapest. ``"highest"`` and
+        ``"smooth"`` are not available in 0.3.0 and earlier.
     optimizer : str, default="Nelder-Mead"
         Passed to ``scipy.optimize.minimize`` for ``dependence="joint"``.
         Derivative-free by design: the orthant likelihood has no convenient

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/src/multivariate_probit/orthant/__init__.py

@@ -5,10 +5,17 @@ On import this binds ``orthant.cdf`` to one of two binaries:
 * a valid key present  -> the paid binary is decrypted in memory and bound;
   full accuracy, unlimited dimension.
 * no key / bad key     -> the plaintext free binary is bound (n <= 3,
-  resolution="low" only) and a one-time notice is printed to stderr.
+  resolution="low" only) and a one-time notice is printed to stderr on the
+  first call.
 
-The key is read from ``$ORTHANT_KEY`` or ``~/.orthant/key``.  Get one at
-https://quantecarlo.com/orthant_key
+The key is read from ``$ORTHANT_KEY`` or ``~/.orthant/key`` at import, or
+passed at any time with ``activate(key)`` (no environment needed, no restart):
+
+    >>> from multivariate_probit import orthant
+    >>> orthant.activate("my-key")
+    'paid'
+
+Get a key at https://quantecarlo.com/orthant_key
 """
 import glob
 import importlib.machinery
@@ -18,7 +25,7 @@ import struct
 import sys
 import tempfile
 
-__all__ = ["cdf"]
+__all__ = ["cdf", "activate"]
 
 _URL = "https://quantecarlo.com/orthant_key"
 _DIR = os.path.dirname(os.path.abspath(__file__))
@@ -26,6 +33,7 @@ _MAGIC = b"ORT1"
 _SCRYPT_N, _SCRYPT_R, _SCRYPT_P = 2 ** 14, 8, 1
 
 tier = "free"  # set to "paid" when the paid binary is bound
+_notice = None  # why we are in free mode; printed once, on the first free call
 
 
 def _read_key():
@@ -52,8 +60,12 @@ def _import_extension(data, modname):
         fd = os.memfd_create(modname)
         try:
             os.write(fd, data)
+            path = "/proc/self/fd/%d" % fd
+            # The path has no .so suffix, so the loader must be given
+            # explicitly; inferred from the path, the spec would be None.
             spec = importlib.util.spec_from_file_location(
-                qualified, "/proc/self/fd/%d" % fd)
+                qualified, path,
+                loader=importlib.machinery.ExtensionFileLoader(qualified, path))
             mod = importlib.util.module_from_spec(spec)
             spec.loader.exec_module(mod)
             return mod
@@ -124,18 +136,55 @@ def _warn(msg):
         pass
 
 
+def _free_cdf(free):
+    """The free binary's cdf, printing the free-mode notice on its first call."""
+    def cdf(*args, **kwargs):
+        global _notice
+        if _notice:
+            _warn(_notice)
+            _notice = None
+        return free(*args, **kwargs)
+    cdf.__doc__ = free.__doc__
+    return cdf
+
+
+def activate(key):
+    """Bind the paid binary with ``key``, at any time; return ``"paid"``.
+
+    Unlike the import-time lookup this never falls back: a wrong key raises
+    ``ValueError`` and a binary that cannot load here raises ``ImportError``,
+    leaving the current binding unchanged.
+    """
+    global tier, _notice
+    if not isinstance(key, str) or not key.strip():
+        raise ValueError("orthant: activate() needs a non-empty key string")
+    try:
+        data = _decrypt_paid(key.strip(), _tag())
+    except ImportError as e:
+        if (e.name or "").startswith("cryptography"):
+            raise ImportError("orthant: activate() needs the cryptography package: "
+                              "pip install cryptography (or multivariate-probit[orthant])",
+                              name=e.name) from None
+        raise
+    except Exception as e:
+        raise ValueError("orthant: key not accepted by this binary (%s). "
+                         "Renew or get a key: %s" % (type(e).__name__, _URL)) from None
+    globals()["cdf"] = _import_extension(data, "_ocore").cdf
+    tier, _notice = "paid", None
+    return tier
+
+
 def _bind():
-    """Pick a binary, bind cdf, and return the tier."""
+    """Pick a binary at import, bind cdf, and return the tier."""
+    global _notice
     tag = _tag()
     key = _read_key()
     if key is not None:
         try:
-            mod = _import_extension(_decrypt_paid(key, tag), "_ocore")
+            data = _decrypt_paid(key, tag)
         except ImportError as e:
-            # Not a key problem: a missing dependency or a binary that won't
-            # load on this platform.  Say so rather than blaming the key.
             notice = ("orthant: key found but the paid binary could not be "
-                      "loaded (%s) -- running in FREE mode (n<=3, "
+                      "decrypted (%s) -- running in FREE mode (n<=3, "
                       "resolution='low').\n" % e)
             if (e.name or "").startswith("cryptography"):
                 notice += "         pip install cryptography\n"
@@ -144,15 +193,24 @@ def _bind():
                       "mode (n<=3, resolution='low').\n"
                       "         Renew or get a key: %s\n" % _URL)
         else:
-            globals()["cdf"] = mod.cdf
-            return "paid"
+            try:
+                globals()["cdf"] = _import_extension(data, "_ocore").cdf
+                return "paid"
+            except Exception as e:
+                # The key was right; loading the binary failed. Say so
+                # rather than blaming the key.
+                notice = ("orthant: key accepted but the paid binary could not "
+                          "be loaded (%s: %s) -- running in FREE mode (n<=3, "
+                          "resolution='low').\n" % (type(e).__name__, e))
     else:
-        notice = ("orthant: FREE mode -- n<=3 and resolution='low' only.\n"
-                  "         For full accuracy and higher dimensions, get a "
-                  "key: %s\n" % _URL)
-
-    globals()["cdf"] = _load_free(tag).cdf
-    _warn(notice)
+        notice = ("orthant: FREE mode -- n<=3 and resolution='low' only. No key "
+                  "in $ORTHANT_KEY or ~/.orthant/key when this module was "
+                  "imported;\n"
+                  "         pass one with orthant.activate(key). Get a key: "
+                  "%s\n" % _URL)
+
+    globals()["cdf"] = _free_cdf(_load_free(tag).cdf)
+    _notice = notice
     return "free"
 
 

{multivariate_probit-0.3.0 → multivariate_probit-0.3.2}/tests/test_evaluators.py

@@ -156,3 +156,56 @@ def test_orthant_with_key_runs_above_three(tmp_path):
         tmp_path=tmp_path,
     )
     assert out.returncode == 0, out.stderr
+
+
+@orthant_built
+def test_orthant_activate_rejects_a_wrong_key_and_stays_free(tmp_path):
+    out = _run_orthant(
+        """
+        from multivariate_probit import orthant
+        assert orthant.tier == "free"
+        try:
+            orthant.activate("not-a-key")
+        except ValueError as exc:
+            print(exc)
+        else:
+            raise SystemExit("no error")
+        assert orthant.tier == "free"
+        eta, Y, corr = problem(3)
+        pattern_prob(eta, Y, corr, evaluator="orthant", resolution="low")
+        """,
+        tmp_path=tmp_path,
+    )
+    assert out.returncode == 0, out.stderr
+    assert "key not accepted" in out.stdout
+    assert out.stderr.count("FREE mode") == 1  # the notice comes once, at first use
+
+
+@orthant_built
+@pytest.mark.skipif(not os.environ.get("ORTHANT_KEY"), reason="needs $ORTHANT_KEY")
+def test_orthant_activate_after_import_without_environment(tmp_path):
+    out = _run_orthant(
+        f"""
+        from multivariate_probit import orthant
+        assert orthant.tier == "free"
+        assert orthant.activate({os.environ["ORTHANT_KEY"]!r}) == "paid"
+        eta, Y, corr = problem(5)
+        got = pattern_prob(eta, Y, corr, evaluator="orthant", resolution="highest")
+        ref = pattern_prob(eta, Y, corr)
+        assert np.abs(got - ref).max() < 1e-2, np.abs(got - ref).max()
+        """,
+        tmp_path=tmp_path,
+    )
+    assert out.returncode == 0, out.stderr
+    assert "FREE mode" not in out.stderr
+
+
+@orthant_built
+def test_orthant_loads_an_extension_from_memory():
+    # The paid binary is loaded from bytes. Where os.memfd_create exists the
+    # path has no .so suffix, which once left the spec without a loader.
+    from multivariate_probit.orthant import _import_extension, _tag
+
+    with open(ORTHANT_DIR / f"_ofree.{_tag()}.so", "rb") as f:
+        mod = _import_extension(f.read(), "_ofree")
+    assert callable(mod.cdf)

multivariate_probit-0.3.0/benchmarks/results.csv

@@ -1,55 +0,0 @@
-sigma,d,N,method,median_err,max_err,s_per_row
-equi,3,100,mvp quadrature,1.3373382605444406e-05,5.564767189014663e-05,6.151222012704238e-05
-equi,3,100,mvp orthant (high),5.520783257679379e-05,0.0001977558612211161,9.499299922026694e-06
-equi,3,100,SciPy default,1.7085000221517266e-06,2.872515295393452e-05,0.0042396432999521495
-equi,3,100,pybhatlib ovus,0.00011229346488104974,0.0008524556947959311,6.468125007813797e-05
-equi,3,100,statsmodels,0.028443441360488153,0.09919722660040897,5.445600254461169e-07
-equi,3,100,reference,0.0,0.0,0.2536748156798421
-equi,5,100,mvp quadrature,2.0314513939401896e-05,0.00011324180994054522,0.03316033348994097
-equi,5,100,mvp orthant (high),8.582073115437003e-05,0.0003006023566725635,1.3224330032244325e-05
-equi,5,100,SciPy default,1.3144313745008795e-06,1.410367111304156e-05,0.0065609772101743145
-equi,5,100,pybhatlib ovus,0.00039917378471633624,0.0019016346577496457,9.364616998936981e-05
-equi,5,100,statsmodels,0.024622620503874038,0.12870241315304662,6.427397602237761e-07
-equi,5,100,reference,0.0,0.0,0.5339089114402304
-equi,8,100,mvp orthant (high),7.170267413460912e-05,0.00047284481673742773,2.604458015412092e-05
-equi,8,100,SciPy default,1.0657093905339088e-06,7.46774341275902e-06,0.013163221169961617
-equi,8,100,pybhatlib ovus,0.0003370945286850363,0.0023083975100191556,0.00014096851024078205
-equi,8,100,statsmodels,0.014453125416131472,0.11538392212916126,1.2416898971423507e-06
-equi,8,100,reference,0.0,0.0,1.4400106923800196
-equi,12,100,mvp orthant (high),5.782104286452053e-05,0.0007241255415409431,4.110162000870332e-05
-equi,12,100,SciPy default,3.4970988831498073e-07,4.815527437126987e-06,0.018762066540075464
-equi,12,100,pybhatlib ovus,0.0002047092197251764,0.0035788660943063266,0.00019040222017792984
-equi,12,100,statsmodels,0.006303004648829415,0.07762653255283464,1.182040141429752e-06
-equi,12,100,reference,0.0,0.0,1.6423227695701643
-equi,16,100,mvp orthant (high),1.5439688762976245e-05,0.0006260612582136282,6.588060990907252e-05
-equi,16,100,SciPy default,7.660225022907471e-08,7.498227887592357e-06,0.026153904990060255
-equi,16,100,pybhatlib ovus,1.6800734753088424e-05,0.0011253768512156778,0.00026122707000467924
-equi,16,100,statsmodels,0.0014514737962008938,0.05355898580397629,2.2758700652047993e-06
-equi,16,100,reference,0.0,0.0,1.1454083026899025
-pairwise,3,100,mvp quadrature,4.975128970532983e-06,8.276622284841828e-05,6.013165984768421e-05
-pairwise,3,100,mvp orthant (high),0.0007572622538304392,0.0028934841345368834,9.238499915227294e-06
-pairwise,3,100,SciPy default,1.0863552671880328e-06,1.1356497155445844e-05,0.006485114450042601
-pairwise,3,100,pybhatlib ovus,0.00037654161082817383,0.004713394298326956,6.38040600460954e-05
-pairwise,3,100,statsmodels,0.037060543222106224,0.15997644750429899,4.1702005546540024e-07
-pairwise,3,100,reference,0.0,0.0,0.24052267621998907
-pairwise,5,100,mvp quadrature,5.607947987679163e-06,6.266772635254991e-05,0.03469973466999363
-pairwise,5,100,mvp orthant (high),0.006131866821483319,0.033843011353965835,1.3460909831337631e-05
-pairwise,5,100,SciPy default,2.0348250380219923e-06,8.649425807485955e-06,0.10544797601993196
-pairwise,5,100,pybhatlib ovus,0.0014663128072703105,0.02764332982515838,9.492035984294489e-05
-pairwise,5,100,statsmodels,0.052292238612919734,0.2549403651447862,5.712499842047691e-07
-pairwise,5,100,reference,0.0,0.0,1.2560263323498657
-pairwise,8,100,mvp orthant (high),0.0047540407934710425,0.037869385076046036,2.4516569974366577e-05
-pairwise,8,100,SciPy default,1.8501012514909876e-06,1.4681123618909275e-05,0.8790659057398443
-pairwise,8,100,pybhatlib ovus,0.0015355099588129813,0.010313908094296734,0.00013880397018510847
-pairwise,8,100,statsmodels,0.024908203618945243,0.1426044668035415,9.262599633075297e-07
-pairwise,8,100,reference,0.0,0.0,4.434727118879964
-pairwise,12,100,mvp orthant (high),0.003427530742914099,0.04306180419633249,3.916209010640159e-05
-pairwise,12,100,SciPy default,1.6616917820439989e-06,2.8483187402228438e-05,1.20114108054986
-pairwise,12,100,pybhatlib ovus,0.0007431598443556342,0.01388904489935694,0.00019705019018147142
-pairwise,12,100,statsmodels,0.014263734107914516,0.1854686635860365,1.1522200657054783e-06
-pairwise,12,100,reference,0.0,0.0,9.518902401560045
-pairwise,16,100,mvp orthant (high),0.00374560158932665,0.045287863431270364,5.415496008936316e-05
-pairwise,16,100,SciPy default,1.8019881021374973e-06,9.824805690655108e-06,1.8181418986897915
-pairwise,16,100,pybhatlib ovus,0.0009786447220386783,0.015062755675490355,0.0002470732299843803
-pairwise,16,100,statsmodels,0.010651175036494289,0.18479811475521019,1.4770901179872453e-06
-pairwise,16,100,reference,0.0,0.0,19.51646036452992

multivariate_probit-0.3.0/docs/benchmarks.md

@@ -1,90 +0,0 @@
-# Benchmarks against other tools
-
-**Task.** Score `P(Y = y | x)`, the probability of the pattern each held-out
-row actually has, from one fitted model. Synthetic data: 6 features, linear
-margins, 3,000 training rows, 100 scored rows per d. The model is
-`MultivariateProbit(inner="linear", dependence="pairwise", random_state=0)`.
-Every column scores that same fitted model (η̂, Σ̂), except statsmodels.
-
-There are two shapes of true Σ:
-
-- **Equicorrelated**, ρ = 0.4: the easy case.
-- **Pairwise-shaped**: a rank-2 factor plus noise, projected to the nearest
-  correlation matrix (smallest eigenvalue about 1e-8), the same construction
-  as [studies/ghk.md](studies/ghk.md). A fitted pairwise Σ̂ lands on that PSD
-  boundary, which is hard for every approximation.
-
-**Reference.** SciPy `multivariate_normal.cdf` with `abseps=1e-7`,
-`releps=1e-6`, one row at a time. Error is the absolute difference from it in
-probability: median over rows, maximum in parentheses. Time is wall-clock per
-row after one warm-up call, on a two-core x86-64 Linux machine.
-
-Reproduce with [benchmarks/benchmarks.ipynb](../benchmarks/benchmarks.ipynb)
-(logic in `benchmarks/bench.py`; raw numbers in `benchmarks/results.csv`).
-The reference values are cached in `benchmarks/ref_*.npz` together with the
-problem they were computed for, so a rerun skips the slow part and a changed
-problem recomputes. The `orthant` column needs a key in `$ORTHANT_KEY` or
-`~/.orthant/key`.
-
-### Equicorrelated Σ (ρ = 0.4)
-
-| d | median p | multivariate-probit `orthant` (`resolution="high"`) | multivariate-probit `quadrature` | SciPy, default tolerances | pybhatlib (Bhat OVUS) | statsmodels |
-| --- | --- | --- | --- | --- | --- | --- |
-| 3 | 0.348 | 9.5e-6 s / 0.0001 (0.0002) | 6.2e-5 s / 0.0000 (0.0001) | 4.2e-3 s / 0.0000 (0.0000) | 6.5e-5 s / 0.0001 (0.0009) | 5.4e-7 s / 0.0284 (0.0992) |
-| 5 | 0.154 | 1.3e-5 s / 0.0001 (0.0003) | 3.3e-2 s / 0.0000 (0.0001) | 6.6e-3 s / 0.0000 (0.0000) | 9.4e-5 s / 0.0004 (0.0019) | 6.4e-7 s / 0.0246 (0.1287) |
-| 8 | 0.101 | 2.6e-5 s / 0.0001 (0.0005) | impractical past d = 7 | 1.3e-2 s / 0.0000 (0.0000) | 1.4e-4 s / 0.0003 (0.0023) | 1.2e-6 s / 0.0145 (0.1154) |
-| 12 | 0.022 | 4.1e-5 s / 0.0001 (0.0007) | impractical past d = 7 | 1.9e-2 s / 0.0000 (0.0000) | 1.9e-4 s / 0.0002 (0.0036) | 1.2e-6 s / 0.0063 (0.0776) |
-| 16 | 0.0024 | 6.6e-5 s / 0.0000 (0.0006) | impractical past d = 7 | 2.6e-2 s / 0.0000 (0.0000) | 2.6e-4 s / 0.0000 (0.0011) | 2.3e-6 s / 0.0015 (0.0536) |
-
-### Pairwise-shaped Σ (near-singular)
-
-| d | median p | multivariate-probit `orthant` (`resolution="high"`) | multivariate-probit `quadrature` | SciPy, default tolerances | pybhatlib (Bhat OVUS) | statsmodels |
-| --- | --- | --- | --- | --- | --- | --- |
-| 3 | 0.431 | 9.2e-6 s / 0.0008 (0.0029) | 6.0e-5 s / 0.0000 (0.0001) | 6.5e-3 s / 0.0000 (0.0000) | 6.4e-5 s / 0.0004 (0.0047) | 4.2e-7 s / 0.0371 (0.1600) |
-| 5 | 0.171 | 1.3e-5 s / 0.0061 (0.0338) | 3.5e-2 s / 0.0000 (0.0001) | 1.1e-1 s / 0.0000 (0.0000) | 9.5e-5 s / 0.0015 (0.0276) | 5.7e-7 s / 0.0523 (0.2549) |
-| 8 | 0.097 | 2.5e-5 s / 0.0048 (0.0379) | impractical past d = 7 | 8.8e-1 s / 0.0000 (0.0000) | 1.4e-4 s / 0.0015 (0.0103) | 9.3e-7 s / 0.0249 (0.1426) |
-| 12 | 0.025 | 3.9e-5 s / 0.0034 (0.0431) | impractical past d = 7 | 1.2e+0 s / 0.0000 (0.0000) | 2.0e-4 s / 0.0007 (0.0139) | 1.2e-6 s / 0.0143 (0.1855) |
-| 16 | 0.013 | 5.4e-5 s / 0.0037 (0.0453) | impractical past d = 7 | 1.8e+0 s / 0.0000 (0.0000) | 2.5e-4 s / 0.0010 (0.0151) | 1.5e-6 s / 0.0107 (0.1848) |
-
-R `mvProbit` was not run (see below).
-
-The reference itself took 0.24 to 1.6 s per row on the equicorrelated Σ and
-0.24 to 20 s per row on the pairwise-shaped one. At d = 16 on the
-equicorrelated Σ most pattern probabilities are small (median 0.0024), so the
-maximum error says more than the median there.
-
-## Reading the tables
-
-- **statsmodels** fits one `Probit` per outcome and has no correlation matrix,
-  so its pattern probability is a product of marginals. Its error is the cost
-  of ignoring Σ, not of numerical integration, and it does not shrink with
-  more data. Its time covers prediction only.
-- **`orthant`** scores all rows in one compiled call. It was the fastest of
-  the integrating methods at every d on both Σ shapes: 4 to 7 times faster
-  than pybhatlib, and 390 to 510 times faster than SciPy on the
-  equicorrelated Σ. On the equicorrelated Σ it also had the smallest maximum
-  error of the approximations. On the pairwise-shaped Σ it was the *least*
-  accurate integrating method: median error 0.0008 to 0.0061, maximum up to
-  0.045.
-- **pybhatlib** scores with Bhat's analytic approximation (`mvncd_batch`,
-  `method="ovus"`), one row at a time in Python. It is 65 to 100 times faster
-  than SciPy on the equicorrelated Σ. On the pairwise-shaped Σ its median error
-  (0.0004 to 0.0015) was 2 to 5 times smaller than `orthant`'s, and so was its
-  maximum from d = 5 up. pybhatlib also has a PyTorch backend that runs on
-  your own GPU, but in pybhatlib 0.4.0 `mvncd` converts its inputs to NumPy
-  and `mvncd_batch` loops over rows in Python, so the GPU does not speed up
-  this scoring.
-- **SciPy** stays accurate on the pairwise-shaped Σ but slows sharply, to
-  0.9 to 1.8 s per row from d = 8. That is 65 to 70 times its time on the
-  equicorrelated Σ at the same d.
-- **`quadrature`** matched the reference on both shapes up to d = 5.
-- **R `mvProbit`** was not run: R is not installed on the benchmark machine,
-  and the notebook has no R column yet.
-
-## What this does not establish
-
-- Two Σ shapes, one seed, and 100 rows per d. The pairwise-shaped Σ is one
-  construction of a near-singular matrix; see [studies/ghk.md](studies/ghk.md)
-  and [studies/comparators.md](studies/comparators.md) for more.
-- Scoring only. Fitting time against full-likelihood estimation is in
-  [studies/comparators.md](studies/comparators.md).